FDI Inequality Europe

FDI and Income Inequality: Evidence from Europe

by Dierk Herzer and Peter Nunnenkamp

No. 1675 | January 2011

Kiel Institute for the World Economy, Hindenburgufer 66, 24105 Kiel, Germany

Kiel Working Paper No. 1675 | January 2011

FDI and Income Inequality: Evidence from Europe

Dierk Herzer and Peter Nunnenkamp

Abstract: This paper examines the relationship between foreign direct investment (FDI) and income inequality for a sample of ten European countries over the period 1980 to 2000. Using panel co-integration and causality techniques that are robust to omitted variables, slope heterogeneity, and endogenous regressors, we find that: (1) FDI has a positive short-run effect on income inequality in Europe, (2) the long-run effect of FDI on inequality, however, is negative on average, (3) long-run causality runs in both directions, suggesting that an increase in FDI reduces income inequality and, in turn, higher inequality leads to lower FDI inflows, and (4) there are large differences in the long-run effect of FDI on income inequality, with two countries (Ireland and Spain) exhibiting a positive relationship between FDI and income inequality.

Keywords: FDI, income inequality, panel co-integration, Europe

JEL classification: F21, D31, C23 Dierk Herzer

Schumpeter School of Business and Economics, University of Wuppertal Gaustr. 20 42119 Wuppertal Germany Phone: 0049-202-439-3006 Fax: 0049-202-439-3011 E-mail: [email protected]

Peter Nunnenkamp Kiel Institute for the World Economy Hindenburgufer 66 24105 Kiel Germany Phone: 0049-431-8814-209 Fax: 0049-0431-8814-500 E-mail: [email protected]

____________________________________ The responsibility for the contents of the working papers rests with the author, not the Institute. Since working papers are of a preliminary nature, it may be useful to contact the author of a particular working paper about results or caveats before referring to, or quoting, a paper. Any comments on working papers should be sent directly to the author. Coverphoto: uni_com on photocase.com

1

1. Introduction

Foreign direct investment (FDI) is often found to have positive effects on growth and

productivity in the host country, but what is generally neglected is the issue of equality (Figini

and Grg 2006: 1). Evidence on the distributional consequences of inward FDI is particularly scarce

in the European context. This is surprising as public concerns about globalisation widening the

income gap between rich and poor segments of the population figure prominently in current

political and economic discourse in various European countries.

Previous studies focused on the labour market repercussions of outward FDI in the home

countries of European companies engaged in international outsourcing and offshoring.

Summarising this literature, Geishecker et al. (2008: 169) conclude that international outsourcing

leads to a shift in the relative demand for labour, which has implications for the wage bill as well

as the employment prospects of high- versus low-skilled labour. However, undesirable effects

upon the labour market (Taylor and Driffield 2005: 243) may also result from European countries

being the target of FDI by companies based within or outside the region. According to the World

Investment Report, Europe accounted for 46 per cent of the worldwide stock of inward FDI in

2009just slightly less than its share of 53 per cent of outward FDI stocks (UNCTAD 2010).

We argue in Section 2 that the effects of inward FDI on income inequality in European host

countries are theoretically ambiguous. Against this background, we explore the relationship

between FDI stocks and income inequality for a sample of 10 European countries over the period

1980 to 2000. We employ panel co-integration and causality techniques that are robust to omitted

variables, slope heterogeneity, and endogenous regressors (Section 3). Our results reported in

Section 4 reveal that inward FDI reduces income inequality, though only in the longer run and not

necessarily in less advanced host countries such as Ireland and Spain. At the same time, less

unequal host countries attract more FDI.

2. Theoretical background and related empirical literature

The European host countries in our sample differ considerably in their overall level of

economic development. The sample includes Malta, Spain, and Ireland, with an average per-capita

income in the range of 3,400-5,200 US$ in 1980 (i.e., at the beginning of our period of observation)

as well as Germany, the Netherlands, Norway, and Sweden, with a per-capita income in the range

of 10,800 and 14,000 US$ (see also Section 3). Arguably, the distributional consequences of FDI

from relatively rich source countries (within and outside the sample) in the poorer host countries in

the sample may be similar to those predicted by North-South models, such as Feenstra and

Hansons (1997). Accordingly, the availability of relatively cheap labour in the poorer host

2

countries may encourage firms based in the richer source countries to undertake cost-oriented,

vertical FDI by offshoring labour-intensive parts of the production process. This type of FDI may

increase the skill premium not only in the richer source country, but also in the poorer host country.

This could happen when offshoring involves activities that are relatively skilled-labour intensive in

the host country, even though they are relatively unskilled-labour intensive in the source country.

Several empirical studies support the hypothesis derived from endowment-driven North-

South models, according to which FDI is associated with greater inequality by raising the skill

premium in poorer host countries. The case of Mexico has received particular attention (Aitken et

al. 1996; Feenstra and Hanson 1997; Hanson 2003). Additionally, FDI has benefited skilled

workers more than unskilled workers in other emerging economies, including Indonesia (Lipsey

and Sjholm 2004) and Korea (Mah 2002). Cross-country studies covering a varying number of

developing countries produced more ambiguous findings (e.g., Tsai 1995; Sylwester 2005; Choi

2006).

North-South FDI models are less useful for analysing the distributional consequences of FDI

in the richer host countries in our sample; vertical FDI in these models tends to result from large

wage differentials between the source and the host country. However, the argument of Markusen

(1995), that advanced countries with similar relative factor endowments host the bulk of FDI,

applies in the European context, too. For instance, just five EU countries accounted for two thirds of

total inward FDI stocks in the EU in 2000, even though their average per-capita income varied by

less than 10 per cent at that time (UNCTAD 2010).1 Markusen (1995) posits that advanced

countries attract mainly market-oriented, horizontal types of FDI. Blonigen and Slaughter argue that

it is quite difficult to assess theoretically whether and how this [type of] inward FDI should affect

wages.

The relevant theoretical literature departs from the observation that multinational enterprises

(MNEs) possess firm-specific assets, such as technological knowledge and management skills,

granting them a productivity advantage over domestic firms. MNEs can then be regarded as

vehicles for transferring new technologies, know-how, and ideas to the host country, acting as

role models for indigenous firms (Figini and Grg 1999: 596). Yet, recent MNE models make

different predictions on the distributional effects of FDI in advanced countries. Assuming that the

operations of MNEs are more skill intensive than the operations of domestic firms, MNEs could

lead to greater income inequality by increasing the demand for skilled labour (see, e.g., Taylor and

Driffield 2005).

1 Belgium, France, Germany, the Netherlands, and the United Kingdom.

3

The opposite conclusion, that FDI results in less inequality, can be reached, however, when

refining the ranking of skill intensities at the firm and plant level. Markusen and Venables (1997)

consider headquarter (HQ) services to be more skill intensive than plant operations. It has also to be

taken into account that advanced countries are typically host, as well as home, to (foreign and,

respectively, domestic) MNEs. Under such conditions, the establishment of foreign plant operations

through FDI may reduce the relative demand for skilled labour in the host country. This is most

likely to happen where foreign plant operations are considerably less skill intensive than the HQ

services supplied from the foreign MNEs home base, and where the HQ services of various

domestic MNEs have traditionally shaped the demand for skilled labour. The United States spring

to mind as a case in point (Blonigen and Slaughter 2001), but advanced European countries are also

likely to fit into this reasoning.

Theoretical predictions on the distributional effects of FDI in advanced host countries

become still more complex when accounting for learning and skill upgrading in the transition to a

new technological paradigm (Aghion and Howitt 1998: 262). While domestic firms may benefit

from FDI-induced spillovers, their absorption of new technologies may increase inequality in the

short run and reduce inequality in the longer run. Aghion and Howitt (1998, chapter 8) model such

a transition by explicitly referring to the Kuznets inverted-U hypothesis of rising and then falling

inequality. Accordingly, the skill premium increases as long as learning efforts result in a high

demand for skills that are in short supply. Subsequently, wage inequality declines to the extent that

the supply of the required skills improves and firms have managed the transition to the new

technological paradigm.

Empirical studies addressing these theoretical predictions are still few. Some country-

specific analyses report ambiguous findings. For the United States, Blonigen and Slaughter (2001)

do not find any evidence that inward FDI contributed to skill upgrading in manufacturing industries

until the mid-1990s. Chintrakarn et al. (2011) perform panel co-integration analysis across the US

states; FDI at the state level reduced income inequality during the period 1977 to 2001, on average,

but the effects proved to be heterogeneous. By contrast, Taylor and Driffield (2005) show for the

United Kingdom that FDI had a positive impact on wage inequality within industries. According to

Figini and Grg (1999), the Irish case reveals an inverted U-shaped pattern, with FDI first

increasing and then later reducing inequality.

In addition to country-specific studies, some indications exist that the distributional effects

of FDI differ between advanced and developing host countries. Gopinath and Chen (2003) analyse

time-series data on FDI and wages for 15 advanced and 11 developing countries. Inward FDI is

associated with higher labour shares in GNP in both sub-samples, but F tests reveal that FDI effects

4

differ significantly between advanced and developing host countries. These authors also report

some evidence that FDI increases the wage gap between skilled and unskilled workers in

developing countries. However, they do not address the wage gap in advanced host countries.

Similarly, Figini and Grg (2006) find that wage inequality increases with FDI in developing host

countries, though this effect diminishes with further increases in FDI. By contrast, inequality

decreases with FDI in advanced host countries.

While focusing on relatively advanced European host countries, the varying level of

economic development within our sample offers more insights into the heterogeneity of FDI effects

on income inequality. Furthermore, the subsequent analysis helps overcome several shortcomings

of almost all previous studies. With the exception of Chintrakarn et al. (2011), earlier analyses are

typically restricted to wage disparity within the manufacturing sector. Including FDI in services is

warranted, and not only because it is quantitatively important; its distributional effects may also

differ from those of FDI in manufacturing, as vertical FDI is more likely to exist in manufacturing

than in services. Finally, as stressed by Lindert and Williamson (2001) and Basu and Guariglia

(2007), wage inequality provides an incomplete picture of income inequality.

3. Empirical model and data

The analysis will examine the relationship between FDI and income inequality in Europe.

This section presents the basic empirical model, discusses some empirical issues, and describes the

data.

3.1. Empirical specification and econometric issues

Following Chintrakarn et al. (2011), we assume that the correct specification of the long-run

relationship between income inequality and FDI is given by a bivariate model of the form

ititiiit GDPFDIbtaInequality +++= )/( , (1) where Ni ...,,2,1= is the country index, Tt ...,,2,1= is the time index, Inequalityit stands for the estimated household income inequality (EHII) in Gini format (measured on a 0 to 100 scale), and

itGDPFDI )/( represents the percentage share of FDI in GDP. As in Figini and Grg (2006) and

Chintrakarn et al. (2011), we use FDI stocks, rather than FDI flows. Stocks should capture long-run

effects more effectively than annual FDI flows, which fluctuate considerably. The coefficient b

represents the long-run effect of FDI on inequality. The ai and it are, respectively, country-specific

fixed effects and country-specific deterministic time trends, capturing any country-specific omitted

factors that are relatively stable or which evolve smoothly over time.

5

An important feature of Equation (1) is the assumption that, in the long-run, permanent

changes in the FDI-to-GDP ratio are associated with permanent changes in the inequality index.

Empirically, this implies that both the individual time series for the EHII Gini coefficient and FDI

(relative to GDP) must exhibit unit-root behaviour, and that Inequalityit must be co-integrated with

itGDPFDI )/( .2 A regression consisting of co-integrated variables has a stationary error term, it,

implying that no relevant integrated variables are omitted; any omitted non-stationary variable that

is part of the co-integrating relationship would enter the error term, thereby producing non-

stationary residuals and a failure to detect co-integration. If there is co-integration between a set of

variables, this same stationary relationship also exists in extended variable space. Thus, an

important implication of finding co-integration is that no relevant integrated variables in the co-

integrating vector are omitted. Co-integration estimators are therefore robust (under co-integration)

to the omission of variables that do not form part of the co-integrating relationship. This justifies a

reduced-form model such as Equation (1) (if co-integrated).

Another assumption inherent in Equation (1) is that inequality is endogenous in the sense

that, in the long run, changes in FDI cause changes in income inequality. However, although the

existence of co-integration implies long-run Granger causality in at least one direction, causality

may also run from inequality to FDI. For example, growing income inequality may reflect a decline

in the real wages of less-skilled workers. Under such conditions, multinationals may undertake

vertical FDI and locate their low-skilled activities in countries with a higher level of inequality in

order to take advantage of lower wages. On the other hand, foreign investors may shy away from

countries with greater levels of income inequality, fearing social conflict and political instability.

Horizontal FDI may also be discouraged to the extent that greater inequality implies limited

effective demand because of a shrinking middle class. It follows that greater inequality could be the

cause of increased or reduced FDI activity. The empirical implication is that, in addition to

examining the time-series properties of the variables and testing whether the variables are co-

integrated, it is also important to deal with this potential endogeneity problem and to investigate the

direction of causality.

2 Strictly speaking, of course, the stochastic process for Gini coefficients cannot be a pure unit-root process. Gini inequality measures are bounded, but we know that a unit-root process will cross any finite bound with probability 1. Nevertheless, as argued by Jones (1995), it may be the case that in the relevant range, such variables are well characterised by a unit-root process. Specifically, if the determining factors of Gini coefficients, such as tastes, time preferences, and government policies, change over time, we observe Gini coefficient series with permanent movements that can be well approximated by a unit-root process. For example, the individual country unit-root tests used by Guest and Swift (2008) suggest that the Gini coefficients for all countries in their study (UK, USA, Australia, Japan, and Sweden) are I(1). Thus, it may well be that inequality measures and FDI-to-GDP ratios are integrated and co-integrated, as the results by Chintrakarn et al. (2011) for the United States suggest.

6

Cross-country heterogeneity represents a further econometric issue to be addressed. As

discussed in Section 2, theoretical arguments suggest that the effect of FDI on inequality may be

highly heterogeneous across countries. Moreover, individual country studies tend to find different

results across regions; the study by Chintrakarn et al. (2011), in particular, suggests that the

inequality effects of FDI are not constant across the US states. Given that traditional homogeneous

panel estimators produce inconsistent and potentially misleading point estimates of the sample

mean of the co-integrating vectors when the true slope coefficients are heterogeneous (see, e.g.,

Pesaran and Smith 1995), we use heterogeneous panel estimators based on the mean group

approach. This also allows us to explicitly analyse the potential heterogeneity in the effects of FDI

across countries.

A final empirical issue is the potential non-linearity in the relationship between FDI and

inequality. As discussed above, the model of Aghion and Howitt (1998) predicts that wage

dispersion first increases after the introduction of new technologies and then decreases (due to

learning processes). This implies that the effect of FDI on inequality may differ, not only from

country to country, but also over time. More specifically, one might expect that the long-run effect

differs significantly from the short-run effect. In our analysis, we therefore attempt to provide

insight into the dynamics of FDI over time by examining both the long-run and short-run effects of

FDI on income inequality.

3.2. Data and descriptive statistics

We now briefly describe the data. The data on itGDPFDI )/( come from UNCTADs FDI

database (available at http://www.unctad.org/templates/Page.asp?IntItemID=3277&lang=1). The

EHII data are taken from the University of Texas Inequality Project (available at

http://utip.gov.utexas.edu/data.html). The major advantage of the EHII source is that the data are

fully comparable across space and time. The EHII data set combines information from the United

Nations Industrial Development Organization (UNIDO) with information from the Deininger and

Squire data set; it also uses other relevant information, such as the ratio of manufacturing

employment to total population, the degree of urbanisation, and population growth.3

3 See Galbraith (2009) for a detailed discussion of the data set and its construction.

7

Table 1

Countries and country summary statistics

Country Average of Inequalityit Average of (FDI/GDP)it Finland 31.72 5.37 Germany 33.51 6.34 Ireland 39.58 112.57 Italy 35.30 5.03 Malta 33.78 23.91 Netherlands 33.02 24.47 Norway 33.00 12.63 Spain 37.13 12.22 Sweden 27.71 9.61 United Kingdom 32.40 17.28

The identification and estimation of co-integrating relationships requires the use of data over

a sufficiently long period of time. Panel co-integration methods can be implemented with shorter

data spans than their time-series counterparts due to exploitation of both the time-series and cross-

sectional dimensions of the data. Consequently, a period of about 20 observations should be long

enough in the present panel context. We include all countries for which continuous data are

available over the period from 1980 to 2000 (21 years), resulting in a balanced panel of 210

observations on 10 countries.

Table 1 lists the countries along with the average values for Inequalityit and itGDPFDI )/(

over the period of observation. Ireland is the country with the highest inequality, followed by Spain

and Italy, while Sweden ranks at the bottom of the inequality scale. Ireland also heads the list of

FDI recipients (relative to GDP), followed by the Netherlands and Malta. The countries with the

lowest FDI-to-GDP ratios are Germany, Finland, and Italy.

4. Empirical analysis

The pre-tests for unit roots and co-integration are reported in the Appendix. They suggest

that the variables are non-stationary and co-integrated, as assumed in Equation (1). In this section,

we provide estimates of the co-integrating relationship between FDI and income inequality. We also

investigate the direction of causality between the two variables, explore the short-run dynamics of

the relationship, and examine the degree of heterogeneity in the effects of FDI on income inequality

across countries.

4.1. Long-run relationship

We estimate the long-run effect of FDI on income inequality using the between-dimension

group-mean panel DOLS estimator that Pedroni (2001) argues has a number of advantages over the

8

within-dimension approach. First, it allows for greater flexibility in the presence of heterogeneous

co-integrating vectors, whereas under the within-dimension approach, the co-integrating vectors are

constrained to being the same for each country. This is an important advantage for applications such

as the present one, as the effect of FDI on income inequality cannot reasonably be assumed to be

the same across countries. Another advantage of the between-dimension estimators is that the point

estimates provide a more useful interpretation in the case of heterogeneous co-integrating vectors;

they can be interpreted as the mean value of the co-integrating vectors, which is not appropriate for

the within estimators. Finally, the between-dimension estimators suffer from much lower small-

sample-size distortions than is the case with the within-dimension estimators.

The DOLS regression is given by

it

p

pjjitijitiiiit

i

i

GDPFDIGDPFDIbtaInequality + +++==

)/()/( , (2)

where ij are coefficients of lead and lag differences, which account for possible serial correlation

and endogeneity of the regressor(s), thus yielding unbiased estimates. Thus, an important feature of

the DOLS procedure is that it generates unbiased estimates for variables that co-integrate, even with

endogenous regressors. In addition, the DOLS estimator is superconsistent under co-integration, and

it is also robust to the omission of variables that do not form part of the co-integrating relationship.

From regression (2), the group-mean DOLS estimator for b is constructed as

11 1

1

1

1 '~'

=

= =

= N

i

T

titit

T

titit szzzNb , (3)

where itz is the 1)1(2 +K vector of regressors itz = [ itGDPFDI )/( ,)/( iGDPFDI ,)/( KitGDPFDI , KitGDPFDI + )/( ], its~ = iit ss , and the subscript 1 outside the brackets

indicates that only the first element of the vector is taken to obtain the pooled slope coefficient.

Because the expression following the summation over the i is identical to the conventional time-

series DOLS estimator, the between-dimension estimator for b is calculated as

=

=N

iibNb

1

1 , (4)

where ib is the conventional DOLS estimator applied to the ith country of the panel and

== Ni bb itNt 1 2/1 is the associated t statistic. As found by Stock and Watson (1993), this estimator performs well in short time series compared to other co-integration estimators such as the maximum

likelihood estimator of Johansen (1988), or the fully modified ordinary least squares (FMOLS)

estimator of Phillips and Hansen (1990).

9

The DOLS group-mean estimate of the long-run relationship is as follows (t statistics in

parentheses; ** and (*) denote significance at the 1% (5%) level):4

)31.3()79.12()59.103()/(**132.0**261.0**604.31

+= itit GDPFDItInequality . (5)

The coefficient on FDI is highly significant and negative, suggesting that FDI reduces income

inequality in European host countries (on average) in the long run. This result is consistent with the

findings of Chintrakarn et al. (2011) for the US states and supports the notion that the effects of FDI

on inequality in advanced countries tend to differ from those in developing countries.

Given the small number of countries in our sample, we need to ensure that the negative

average effect on inequality is not due to outliers. To this end, we re-estimate the DOLS regression,

excluding one country at a time from the sample. The sequentially estimated coefficients and their

t statistics are presented in Figure 1. They fluctuate between -0.29 (due to the exclusion of Spain)5

and -0.08 and are always significant at least at the 5-per-cent level, suggesting that the negative

effect of FDI on income inequality is not the result of possible outliers.

Figure 1

DOLS estimation with single country excluded from the sample

4.2. Short-run and long-run causality

We are interested in detecting the direction of long-run causality, even though estimation by

DOLS does not require the regressor(s) to be exogenous (and co-integration implies long-run

Granger causality in at least one direction). Specifically, the arguments in Section 3.1 suggest that

causality may run in both directions, that is, not only from FDI to inequality but also from

4 The number of leads and lags in the individual DOLS regressions was determined by the Schwarz criterion with a maximum of two lags. 5 The individual country DOLS estimates are presented in Table 3.

t statistics of the coefficients

-4.8

-4.4

-4.0

-3.6

-3.2

-2.8

-2.4

-2.0

1 2 3 4 5 6 7 8 9 10

Coefficients on (FDI/GDP)it

-.32

-.28

-.24

-.20

-.16

-.12

-.08

1 2 3 4 5 6 7 8 9 10

10

inequality to FDI. To test the direction of long-run causality and to examine the short-run dynamics

between the variables, we enter the residuals from the individual DOLS long-run relations,

])/([ itiiiitit GDPFDIbtaInequalityec ++= , (6) as error-correction terms into a simple panel vector-error-correction model (VECM) of the form

+

+

+

=

= it

itit

jit

jitk

jj

i

i

it

it ecaa

GDPFDIInequality

cc

GDPFDIInequality

2

11

2

1

12

1

)/(

)

)/()

, (7)

where the cis are fixed effects; the error-correction term, 1itec , represents the error in, or deviation

from, the equilibrium; and the adjustment coefficients 1a and 2a capture how Inequalityit and

itGDPFDI )/( respond to deviations from the equilibrium relationship. From the Granger

representation theorem, we know that at least one of the adjustment coefficients must be non-zero if

a long-run relationship between the variables is to hold. A significant error-correction term also

suggests long-run Granger causality and thus, long-run endogeneity (see, e.g., Hall and Milne

1994), whereas a non-significant adjustment coefficient implies long-run Granger non-causality

from the independent to the dependent variable(s), as well as weak exogeneity. Following Herzer

(2008), we test for weak exogeneity by first eliminating the insignificant short-run dynamics in the

model successively, according to the lowest t values, and then deciding on the significance of the

error-correction term. In doing so, we reduce the number of parameters and thereby increase the

precision of the weak exogeneity tests on the coefficients. Since all variables in Equation 7,

including 1itec , are stationary (because the level variables are integrated of order 1 and co-

integrated), conventional t tests can be used for this purpose.

Table 2

Vector-error-correction model, long-run causality, and short-run dynamics

Independent variables Dependent variable Inequalityit

Dependent variable (FDI/GDP)it

1itec -0.419** (-6.45) 0.792* (2.37) Inequalityit-1 -0.131* (-2.00) Inequalityit-2 (FDI/GDP)it-1 0.040* (2.19) 0.399** (3.77) (FDI/GDP)it-2 0.223* (2.07) Notes: ** and (*) indicate significance at the 1% (5%) level. t statistics are in parentheses. The number of lags was determined by the general-to-specific procedure with a maximum of two lags. Given that the insignificant short-run dynamics were eliminated successively according to the lowest t values, they are not reported here.

Table 2 presents the results. As the t statistics for the error-correction terms indicate, the null

hypothesis of weak exogeneity can be rejected for both Inequalityit and (FDI/GDP)it, at least at the

11

5-per-cent level. Both income inequality and FDI are thus endogenous. From this it can be

concluded that the statistical long-run causality is bidirectional, in turn suggesting that decreased

inequality is both a consequence and a cause of increased FDI.

Another important result is that the coefficient on (FDI/GDP)it-1 is statistically significant

and positive in the equation with Inequalityit as the dependent variable (Table 2, second column).

This indicates that FDI has a positive causal effect on income inequality in the short run, while its

long-run effect on income inequality is negative.

To test the robustness of this conclusion, we compute the impulse-response functions from

the VECM residuals, applying a standard Choleski decomposition (with Inequalityit ordered first

and (FDI/GDP)it ordered last). The responses of income inequality to a one-standard-deviation

innovation in FDI over a 10-year horizon are shown in Figure 2, where the dashed lines mark plus

and minus two standard errors obtained through Monte Carlo simulations using 1,000 replications.

The figure reveals that a one-standard-deviation innovation in FDI first produces an increase in

income inequality, reaching a peak in two years. After four years, however, income inequality

gradually and permanently decreases in response to a one-standard-deviation innovation in the FDI-

to-GDP ratio. Consequently, the impulse responses are consistent with the above finding that FDI

has a positive effect on inequality in the short run, whereas the long-run effect of FDI on inequality

is negative.

Figure 2. Response of inequality to FDI

4.4. Individual country effects The results reported so far indicate that FDI has, on average, a negative long-run effect on

income inequality in Europe (and vice versa). This finding for the sample as a whole does not

-.4

-.2

.0

.2

.4

1 2 3 4 5 6 7 8 9 10

12

necessarily imply, however, that FDI affects inequality negatively in each individual country. Table

3 (second and third columns) presents the individual DOLS country estimates of the coefficients on

(FDI/GDP)it. These estimates should be interpreted with caution, given the relatively short time

dimension of our data. However, several previous co-integration analyses for individual countries

are based on smaller time samples (see, e.g., de Crombrugghe et al. 1997; Irvin and Izurieta 2000).

More importantly, we also report the results of the Johansen (1988) maximum likelihood estimator

(fourth and fifth columns) to ensure that our conclusions are robust to different estimation

techniques. Note that the Johansen procedure corroborates that FDI has, on average, a negative

long-run effect on income inequality.

At the same time, there appears to be considerable heterogeneity in the long-run effects of

FDI on income inequality across countries. Specifically, we find that in eight countriesFinland,

Germany, Italy, Malta, the Netherlands, Norway, Sweden, and the United Kingdoman increase in

FDI is associated with a decrease in income inequality, while in two casesIreland and Spainan

increase in FDI is associated with an increase in inequality. This heterogeneity appears to be largely

in line with the reasoning in Section 2. Accordingly, countries with relatively low per-capita

income, such as Ireland and Spain, are most likely to attract vertical FDI.6 The prediction of North-

South models, that inequality widens with FDI, should thus be relevant for these cases, in contrast

to the larger part of our sample, which is characterised by higher per-capita incomes and similar

factor endowments. Table 3

Individual country estimates

DOLS estimator Johansen maximum-likelihood

estimator (FDI/GDP)it t stat. (FDI/GDP)it t stat. Finland -0.576* -2.511 -0.427** -5.107 Germany -0.351 -0.677 -1.616** -4.683 Ireland 0.061** 3.760 0.051** 3.345 Italy -0.156 -0.178 -1.398+ -1.900 Malta -0.174** -3.041 -0.011 -0.292 Netherlands -0.498** -4.137 -0.362** -4.267 Norway -0.204 -0.928 -3.421** -5.724 Spain 1.346* 2.507 3.990** 5.248 Sweden -0.481** -3.867 -0.298** -8.382 United Kingdom -0.287 -1.387 -0.111** -7.939 Average -0.132 -0.360

Notes: **, (*), and [+] indicate significance at the 1% (5%) [10%] level

6 Recall that Malta represents the third country in our sample with relatively low per-capita income. Unlike Ireland and Spain, however, Malta is not known for having attracted considerable FDI in manufacturing, where vertical FDI figures most prominently. FDI in Malta is probably concentrated in tourism.

13

5. Conclusion

Trade-offs associated with FDI have contributed to waning public support of economic

globalisation. Concerns are primarily about outward FDI helping firm-level competitiveness at the

expense of wage pressure or unemployment of low-skilled workers. Inward FDI may involve

similar distributional conflicts. This would be the case if productivity-enhancing spillovers went

along with widening inequality due to shifts in the relative demand for labour towards higher skills.

In Europe, possible trade-offs between average income gains and greater inequality typically rank

high on the political agenda. However, strikingly different predictions can be derived from recent

theoretical models, and reliable empirical evidence on the distributional effects of FDI is largely

lacking. This applies to inward FDI in particular.

The present study contributes to filling this gap. We examined the macroeconomic

relationship between FDI and income inequality for a sample of 10 European host countries over

the period 1980 to 2000 by employing panel co-integration and causality techniques that are robust

to omitted variables, slope heterogeneity, and endogenous regressors. Our results suggest that: (1)

FDI has a positive short-run effect on income inequality in Europe, (2) the long-run effect of FDI on

inequality, however, is negative on average, (3) long-run causality runs in both directions,

suggesting that an increase in FDI reduces income inequality and that, in turn, higher inequality

leads to lower FDI inflows, and (4) there are large differences in the long-run effect of FDI on

income inequality, with Ireland and Spain exhibiting a positive relationship between FDI in income

inequality.

On average, the European experience leads to the conclusion that policymakers competing

for inward FDI need not be concerned that productivity and growth promoting access to superior

technology and know-how generally deepens the divide between rich and poor segments of the

workforce. Possible trade-offs typically appear to be short-lived. In the longer term, relatively few

European host countries run the risk of giving rise to social conflict by attracting FDI. Yet this

heterogeneity deserves to be analysed in more detail. In particular, country-specific studies may

offer deeper insight into the varying effects different types of FDI may have on income inequality.

Distinguishing between vertical and horizontal FDI provides one possible avenue for further

research, but the distributional effects may also vary between FDI from different sources and

depend on the local environment in the host country, including labour market conditions.

14

Appendix: Panel unit root and cointegration tests

A.1. Panel unit-root tests

To examine the time-series properties of the data, we use the between-dimension panel unit

root test of Im, Pesaran, and Shin (2003) (IPS), which is based on the augmented Dickey-Fuller

(ADF) regression for the ith cross-section unit:

it

p

jjitijitiitit

i

xxzx +++= =

1

1' , (A.1)

where pi is the lag order and zit represents deterministic terms, such as fixed effects or fixed effects

combined with individual time trends. The IPS test tests the null hypothesis of a unit root for all i,

0:0 =iH , against the alternative of (trend) stationarity, 0:1

15

averages of the variables. Accordingly, the cross-sectionally augmented ADF (CADF) regression is

given by

it

p

jjtijti

p

jjitijitiitit vxxxxzx

ii +++++= =

=

0

11

1' , (A.3)

where tx is the cross-section mean of xit, tx = = Ni itxN 11 . The cross-sectionally augmented IPS statistic is the simple average of the individual CADF statistics and is defined as

CIPS = t-bar = =

iN

iitN

1

1 , (A.4)

where it is the OLS t ratio of i in Equation A.4. The corresponding critical values are given by Pesaran (2007).

The results for the variables in levels and in first differences are presented in Table A1.

While neither test rejects the null hypothesis of a unit root in levels, the unit-root hypothesis is

rejected for the first differences. We therefore conclude that Inequalityit and (FDI/GDP)it are

integrated of order 1 (the necessary condition for co-integration in a bivariate context).

A.2. Panel co-integration tests

We use several panel co-integration tests to determine whether there is a long-run

relationship between FDI and income inequality. The first is the two-step residual-based procedure

suggested by Pedroni (1999, 2004), which can be intuitively described as follows. In the first step,

the hypothesised co-integrating regression

ititiiiit GDPFDIbtaInequality +++= )/( (A.5) is estimated separately for each country, thus allowing for heterogeneous co-integrating vectors. In

the second step, the residuals from these regressions, it , are tested for stationarity. To test the null hypothesis of non-stationarity (or no co-integration), Pedroni proposes seven statistics. Four of

these test statistics pool the autoregressive coefficients across different countries during the unit-

root test and thus constrain the autoregressive parameters to be homogeneous across countries.

Pedroni refers to these within-dimension-based statistics as panel co-integration statistics. The other

three test statistics are based upon estimators that average the individually estimated autoregressive

coefficients for each country, thus allowing the autoregressive coefficient to be heterogeneous

across countries. Pedroni refers to these between-dimension statistics as group-mean panel co-

integration statistics. The first of the panel co-integration statistics is a non-parametric variance ratio

test. The second and the third are panel versions of the Phillips and Perron (PP) rho statistic and t

16

statistic, respectively. The fourth statistic is a panel ADF t test analogous to the Levin et al. (2002)

panel unit-root test. Similarly, the first of the group-mean panel co-integration statistics is

analogous to the PP rho statistic, the second is a panel version of the PP t statistic, and the third is a

group mean ADF t test analogous to the IPS (2003) panel unit-root test (described above). The

standardised distributions for the panel and group statistics are given by

)1,0(Nv

N = , (A.6)

where is the respective panel or group statistic, and and are the expected mean and variance of the corresponding statistic, tabulated by Pedroni (1999).

A potential problem with the Pedroni approach is that it does not take into account potential

error cross-sectional dependence, which could bias the results. To test for co-integration in the

presence of possible cross-sectional dependence, we use the two-step procedure suggested by Holly

et al. (2010). In the first step, we apply the common correlated effects (CCE) estimator of Pesaran

(2006) to the static co-integrating regression. Like the cross-sectionally augmented IPS test, the

CCE estimator allows for cross-sectional dependencies that potentially arise from multiple

unobserved common factors and permits the individual responses to these factors to differ across

countries. The cross-sectionally augmented co-integrating regression for the ith cross-section is

given by

ittitiitiiiit eGDPFDIgInequalitygGDPFDIbtaInequality +++++= )/()/( 21 , (A.7) where the cross-sectional averages tInequality = Ni itInequalityN 1 and tGDPFDI )/( =

Ni itGDPFDIN )/(1 serve as proxies for the unobserved factors. In the second step, we compute the cross-sectionally augmented IPS statistic for the residuals from the individual CCE long-run

relations, itiiit GDPFDIbtInequality )/( = , including an intercept. In doing so, we account for unobserved common factors that could be correlated with the observed regressors in both steps.

Residual-based (panel) co-integration tests, such as the ones just described, restrict the long-

run elasticities to being equal to the short-run elasticites. If this restriction is invalid, residual-based

(panel) co-integration tests may suffer from low power (see, e.g., Westerlund, 2007). Another

drawback of residual-based (panel) co-integration tests is that they are generally not invariant to the

normalisation of the co-integrating regression. Therefore, we also use the Larsson et al. (2001)

procedure, which is based on Johansens (1988) maximum likelihood estimation procedure. Like

the Johansen time-series co-integration test, the Larsson et al. panel test treats all variables as

potentially endogenous, thus avoiding the normalisation problems inherent in residual-based co-

17

integration tests. In addition, the Larsson et al. procedure allows the long-run elasticities to differ

from the short-run elasticities and hence does not impose a possibly invalid common factor

restriction. It involves estimating the Johansen vector-error-correction model for each country

separately:

itiitkitik

iikitiit zyyy +++= = 11 , (A.8)

where ity is a p 1 vector of endogenous variables ( ity = [ itInequality , itGDPFDI )/( ]'; p is the

number of variables), and i is the long-run matrix of order p p. If i is of reduced rank, ir < p, it is possible to let iii = , where i is a p ir matrix, the ir columns of which represent the co-integrating vectors, and i is a p ir matrix having p rows which represent the error-correction coefficients. The null hypothesis is that all of the N countries in the panel have a common

co-integrating rank, i.e., at most r (possibly heterogeneous) co-integrating relationships among the p

variables: rrrankH ii = )(:0 for all Ni ,...,1= . The alternative hypothesis is that all the cross-sections have a higher rank: prankH i = )(:1 for all Ni ,...,1= . To test 0H against 1H , a panel co-integration rank trace-test statistic is computed by calculating

the average of the individual trace statistics, })()({ pHrHLRiT :

})()({ pHrHLR NT = =

N

iiT pHrHLRN 1

})()({1 , (A.9)

and then standardising it as follows:

( ))1,0(

)()(})()({

})()({ NZVar

ZEpHrHLRNpHrH

k

kNTLR

= . (A.10)

The mean )( kZE and variance )( kZVar of the asymptotic trace statistic are tabulated by Breitung

(2005) for the model (with an intercept and a trend) we use.

18

Table A.2

Panel co-integration tests

Pedroni (1999, 2004) cointegration test statistics Panel statistic 2.12* Panel PP rho statistic -3.21** Panel PP t statistic -5.49** Panel ADF t statistic -5.45** Group PP rho statistic -1.26 Group PP t statistic -6.07** Group ADF t statistic -6.21**

CIPS statistic for the residuals of the CCE long-run relations -2.75** Co-integration rank r = 0 r = 1 Panel trace statistics 3.58** 0.06 Fisher statistics 37.40* 19.75 Notes: ** and (*) indicate a rejection of the null of no co-integration at the 1% (5%) level. Under the alternative hypothesis, the panel statistic diverges to positive infinity so that the right tail of the normal distribution is used to reject the null hypothesis. The relevant 1% critical value for the CIPS statistic is -2.60. The number of lags was determined by the Schwarz criterion with a maximum of two lags. The Fisher statistic is distributed as 2 with 2N degrees of freedom. It has a critical value of 37.57 (31.41) at the 1% (5%) level.

In addition, we compute the Fisher statistic proposed by Madalla and Wu (1999), which is

defined as

= Ni

ip )log(2 , (A.11)

where pi is the p value of the trace statistic for country i, calculated from the response surface

estimates in MacKinnon et al. (1999). The Fisher statistic is distributed as 2 with 2N degrees of

freedom.

The results of these tests are presented in Table A2. Six of the seven Pedroni statistics reject

the null of no co-integration at conventional levels. Specifically, the ADF-type tests reject the null

hypothesis at the 1-per-cent level. Given that these tests have been shown to have the highest power

for smaller sample sizes, such as T = 21 (see, e.g., Pedroni 2004), the ADF test results, in particular,

provide strong evidence of co-integration. This conclusion is supported by the CIPS, the panel

trace, and the Fisher statistics, which show that itInequality and itGDPFDI )/( are co-integrated (or

exhibit a single co-integrating vector).

19

References

Aghion, P., and P. Howitt. Endogenous Growth Theory. Cambridge, MA: MIT Press, 1998.

Aitken, B., A. Harrison and R.E. Lipsey. Wages and Foreign Ownership: A Comparative Study of

Mexico, Venezuela, and the United States. Journal of International Economics, 40, 1996,

345371.

Basu, P., and A. Guariglia. Foreign Direct Investment, Inequality, and Growth. Journal of

Macroeconomics, 29, 2007, 824839.

Blonigen, B.A., and M.J. Slaughter. Foreign-Affiliate Activity and U.S. Skill Upgrading. Review

of Economics and Statistics, 83, 2001, 362376.

Breitung, J. A Parametric Approach to the Estimation of Cointegration Vectors in Panel Data.

Econometric Reviews, 24, 2005, 151173.

Chintrakarn, P., D. Herzer and P. Nunnenkamp. FDI and Income Inequality: Evidence from a

Panel of US States. Forthcoming in Economic Inquiry, 2011.

Choi, C. Does Foreign Direct Investment Affect Domestic Income Inequality? Applied

Economics Letters, 13, 2006, 811814.

de Crombrugghe, D., F.-C. Palm and J.-P. Urbain. Statistical Demand Functions for Food in the

USA and the Netherlands. Journal of Applied Econometrics, 12, 1997, 615637.

Feenstra, R.C., and G.H. Hanson. Foreign Direct Investment and Relative Wages: Evidence from

Mexicos Maquiladoras. Journal of International Economics, 42, 1997, 371393.

Figini, P., and H. Grg. Multinational Companies and Wage Inequality in the Host Country: The

Case of Ireland. Review of World Economics (Weltwirtschaftliches Archiv), 135, 1999,

594612.

Figini, P., and H. Grg. Does Foreign Direct Investment Affect Wage Inequality? An Empirical

Investigation. Institute for the Study of Labor (IZA), Discussion Paper 2336. Bonn, 2006.

Galbraith, J.K. Inequality, Unemployment and Growth: New Measures for Old Controversies.

Journal of Economic Inequality, 7, 2009, 189206.

Geishecker, I., H. Grg and S. Maioli. The Labour Market Impact of International Outsourcing.

In: D. Greenaway (ed.), Globalization and Labour Market Adjustment. Basingstoke:

Palgrave Macmillan, 2008, 152-173.

Gopinath, M., and W. Chen. Foreign Direct Investment and Wages: A Cross-Country Analysis.

Journal of International Trade and Economic Development, 12, 2003, 285309.

Guest, R., and R. Swift. Fertility, Income Inequality, and Labor Productivity. Oxford Economic

Papers, 60, 2008, 97618.

20

Hall, S.G., and A. Milne. The Relevance of P-Star Analysis to UK Monetary Policy. Economic

Journal, 104, 1994, 597604.

Hanson, G.H. What Has Happened to Wages in Mexico since NAFTA? Implications for

Hemispheric Free Trade. National Bureau of Economic Research, NBER Working Paper

9563. Cambridge, MA, 2003.

Herzer, D. The Long-Run Relationship between Outward FDI and Domestic Output: Evidence

from Panel Data. Economics Letters, 100, 2008, 146149.

Holly, S., M.H. Pesaran and T. Yamagata. A Spatio-Temporal Model of House Prices in the

USA. Journal of Econometrics, 158, 2010, 160173.

Im, K.S., M.H. Pesaran and Y. Shin. Testing for Unit Roots in Heterogeneous Panels. Journal of

Econometrics, 115, 2003, 5374.

Irvin, G., and A. Izurieta. Will the Growing Trade Gap Sink Viet Nam? Some Exploratory

Econometrics. Journal of International Development, 12, 2000, 169186.

Johansen, S. Statistical Analysis of Cointegrating Vectors. Journal of Economic Dynamics and

Control, 12, 1988, 231254.

Jones, C. Time Series Tests of Endogenous Growth Models. Quarterly Journal of Economics,

110, 1995, 495525.

Larsson, R., J. Lyhagen and M. Lthegren. Likelihood-Based Cointegration Tests in

Heterogeneous Panels. Econometrics Journal, 4, 2001, 109142.

Levin, A., C.-F. Lin and C.-S.J. Chu. Unit Root Tests in Panel Data: Asymptotic and Finite-

Sample Properties. Journal of Econometrics, 108, 2002, 124.

Lindert, P.H., and J.G. Williamson. Does Globalization Make the World More Unequal? National

Bureau of Economic Research, NBER Working Paper 8228. Cambridge, MA, 2001.

Lipsey, R.E., and F. Sjholm. Foreign Direct Investment, Education and Wages in Indonesian

Manufacturing. Journal of Development Economics, 73, 2004, 415422.

MacKinnon, J.G., A.A. Haug and L. Michelis. Numerical Distribution Functions of Likelihood

Ratio Tests for Cointegration. Journal of Applied Econometrics, 14, 1999, 563577.

Madalla, G.S., and S. Wu. A Comparative Study of Unit Root Tests with Panel Data and a New

Simple Test. Oxford Bulletin of Economics and Statistics, 61, 1999, 631652.

Mah, J.S. The Impact of Globalization on Income Distribution: The Korean Experience. Applied

Economics Letters, 9, 2002, 10071009.

Markusen, J.R. The Boundaries of Multinational Enterprises and the Theory of International

Trade. Journal of Economic Perspectives, 9, 1995, 169189.

21

Markusen, J.R., and A.J. Venables. The Role of Multinational Firms in the Wage-Gap Debate.

Review of International Economics, 5, 1997, 435451.

Pedroni, P. Critical Values for Cointegration Tests in Heterogeneous Panels with Multiple

Regressors. Oxford Bulletin of Economics and Statistics, 61 (special issue), 1999, 653670.

Pedroni, P. Purchasing Power Parity Tests in Cointegrated Panels. Review of Economics and

Statistics, 83, 2001, 727731.

Pedroni, P. Panel Cointegration: Asymptotic and Finite Sample Properties of Pooled Time Series

Tests with an Application to the PPP Hypothesis. Econometric Theory, 20, 2004, 597625.

Pesaran, M.H. Estimation and Inference in Large Heterogeneous Panels with a Multifactor Error

Structure. Econometrica, 74, 2006, 9671012.

Pesaran, M.H. A Simple Panel Unit Root Test in the Presence of Cross-Section Dependence.

Journal of Applied Econometrics, 22, 2007, 265312.

Pesaran, M.H., and R. Smith. Estimating Long-Run Relationships from Dynamic Heterogeneous

Panels. Journal of Econometrics, 68, 1995, 79113.

Phillips, P.C.B, and B.E. Hansen. Statistical Inference in Instrumental-Variables Regression With

I(1) Processes. Review of Economic Studies, 57, 1990, 99125.

Stock, J.H., and M.W. Watson. A Simple Estimator of Cointegrating Vectors in Higher-Order

Integrated Systems. Econometrica, 61, 1993, 783820.

Sylwester, K. Foreign Direct Investment, Growth and Income Inequality in Less Developed

Countries. International Review of Applied Economics, 19, 2005, 289300.

Taylor, K., and N. Driffield. Wage Inequality and the Role of Multinationals: Evidence from UK

Panel Data. Labor Economics, 12, 2005, 223249.

Tsai, P.-L. Foreign Direct Investment and Income Inequality: Further Evidence. World

Development, 23, 1995, 469483.

UNCTAD. World Investment Report 2010. New York: United Nations, 2010.

Westerlund, J. Testing for Error Correction in Panel Data. Oxford Bulletin of Economics and

Statistics, 69, 2007, 709748.